Migration in Response to Civil Conflict: Evidence from the Border of the American Civil War⇤ Shari Eli University of Toronto and NBER

Laura Salisbury York University and NBER

Allison Shertzer University of Pittsburgh and NBER February 2016

Abstract This paper documents a migration response to the American Civil War. We compare men who served in the Confederate Army with their men who served in the Union army in the border state of Kentucky, which contributed significant numbers of soldiers to both armies. To create the dataset, we collected the universe of existing Union and Confederate enlistees from Kentucky and matched men to their pre- and post-war occupations and place of residence using the 1860 and 1880 censuses. Our findings show that Confederate soldiers were positively selected from the Kentucky population prior to the onset of the conflict. We demonstrate strikingly di↵erent postwar migration patterns between Union and Confederate veterans, and we argue that this is driven by geographic di↵erences in the social returns to having served on each side. Our results suggest that the decision to serve on the Union or Confederate side created lasting social divisions between otherwise similar men, and that these divisions had real economic consequences.

⇤

This is an extremely preliminary and incomplete draft. Please do not cite.

1

1

Introduction

More than half of the nations around the world have faced armed conflicts or civil wars during the last fifty years. Recent literature has shown that civil war is linked to low per capita incomes and slow economic growth.1 While the economic development literature on civil war has increased substantially in the last decade, little is known about the long-term economic e↵ects of armed internal conflict. Moreover, the consequences for individual soldiers of serving on opposing sides of a civil conflict are not well studied. War creates victors and losers, and social divisions between veterans from opposing sides may lead to persistent inequality and lack of economic integration among regions of a country. In this paper, we consider the American Civil War (1861 - 1865) and compare the post-war migration decisions of men who selected into the Confederate Army with their counterparts on the Union side. We focus our analysis on the border state of Kentucky, which contributed numerous soldiers to both armies. The literature on the consequences of civil conflict focuses on the e↵ects of exposure to war rather than the e↵ects of participating on a winning or losing side. In particular, this literature explores the economic consequences of destruction of physical infrastructure, the e↵ects of exposure to violence and disease on later human capital acquisition, and the e↵ects of conflict on institutional development.2 The American Civil War provides a unique context for studying the outcomes of winners and losers in armed civil conflicts. The economic and human costs associated with this war were staggering: hundreds of thousands of men died in the war, and the war imposed billions of (1860) dollars in direct economic costs (Goldin and Lewis 1975). Moreover, sufficient time has passed to observe the later life outcomes of Civil War veterans, which stands in contrast to modern civil conflicts. We focus on documenting a migration response to the Civil War. Economists typically model migration as an investment: individuals migrate in order to maximize their expected lifetime earnings net of mobility costs. In the case of civil conflict, animosity between former combatants may generate migrations which would never have happened for economic reasons in the absence of the conflict. In other words, by imposing social costs on participants, civil conflict leads to “inefficient” migration behavior. This is yet another potential cost associated with civil war. On the other 1

See Blattman and Miguel (2010) for a review of recent literature. Miguel and Roland (2011) measure the e↵ect of exposure to bombing during the Vietnam war on human capital attainment; Bundervoet et al (2009) look at the e↵ect of exposure to conflict in Burundi on child health; Blattman and Annan (2010) measure the e↵ects of being conscripted into military service in Uganda on human capital attainment and income in later life. For a more complete survey of this literature, see Blattman and Miguel (2010). 2

2

hand, conditional on a civil conflict having occurred, migration may reduce prolonged exposure to enemy combatants, and may thus limit ongoing violence. In any case, migration responses have important potential implications about the lasting impacts of civil war. In the case of the American Civil War, migration responses are difficult to meausure. For one thing, most combatants from opposing sides didn’t live anywhere near one another. Most Union soldiers came from the North, and most Confederate soldiers came from the South. Looking at aggregate census data, it appears that the fraction of southern-born individuals living outside the South declined during the years after the Civil War (see figure 1), which suggests that migration from South to North might have slowed or reversed after the war. However, it is difficult to draw causal inferences from this picture. Border states like Kentucky are much more useful in this regard, as soldiers from the Union and Confederate armies came from similar localities, which creates more scope for a measurable migration response. We construct a longitudinal database of Union and Confederate recruits from Kentucky. We do this by matching Union and Confederate military records from Kentucky regiments to the census of 1860, and then linking recruits forward to the census of 1880. This allows us to measure and control for selection into each army on the basis of observable socioeconomic status. We are also able to observe recruits’ county of residence prior to enlistment, which allows us to infer whether recruits were living in places where Union or Confederate status would have been more socially rewarded. Thus, we can determine whether or not Union recruits were “pushed” out of counties that were more socially aligned with the Confederacy, and whether or not they were “pulled” toward counties that were more socially aligned with the Union. The longitudinal nature of our database allows us to deal with concerns about di↵erent migration behavior being driven by di↵erences in skill and not by social rewards or penalties from military service. For example, suppose that Union recruits are systematically less skilled, and that they are more likely to leave counties more sympathetic to the Confederacy. In principle, this could happen because counties more sympathetic to the Confederacy also have higher returns to skill. As such, the ability to observe ex ante characteristics of recruits – especially things like occupational attainment and wealth – is a major advantage of our research design. We find that Union and Confederate recruits from Kentucky displayed very di↵erent migration behavior after the Civil War. We measure social alignment to the Confederacy using 1860 election returns, as well as the prevalence of slavery. We find that Union veterans from more “Confederate” counties were significantly more likely to migrate, and this cannot be explained by di↵erence in 3

average skill. Moreover, conditional on migrating, Union veterans were more likely to opt for places more sympathetic to the Union. We also find that Union and Confederate migrants were di↵erently selected in terms of skill, in a way that is consistent with social rewards or penalties from military service disproportionately a↵ecting skilled workers. This paper is organized in the following way: Section 2 provides a background on Union and Confederate soldiers as well as a review of the literature on the economic and social costs and outcomes of the Civil War. Section 3 provides a discussion of data gathered; Section 4 explains the empirical strategy; Section 5 presents and then discusses our results; and Section 6 concludes.

2

Historical Background

The Civil War began on April 12, 1861 when Confederate ships attacked the Union Army at Fort Sumter, South Carolina and ended on April 9, 1865 when Robert E. Lee surrendered at Appomattox Courthouse in Virginia. Approximately 2.2 million men served on the side of the Union (North) and 1.1 million men served for the Confederacy (South).

2.1

Kentucky during the War

During the Civil War, Kentucky – a border and slave state – did not declare secession from the Union. As in other border states, pro-Confederate and pro-Union supporters lived alongside each other (both Union President Abraham Lincoln and Confederate President Je↵erson Davis were born in Kentucky). Tobacco, whiskey, snu↵ and flour produced in Kentucky were exported to the South and Europe via the Ohio and Mississippi rivers and to the North by rail. Therefore, Kentucky’s economy relied on markets in the Union and the Confederacy. In addition, though most Kentuckians owned no slaves, others were heavily involved in the profitable exportation of slaves to the Deep South. Therefore, since allegiance of Kentuckians was divided for economic and geographic reasons, the state legislature attempted to remain neutral for the first year of the conflict. Remaining neutral, however, proved impossible when the Confederate army invaded the state in the fall of 1861. While the state legislature remained officially loyal to the Union side throughout the war, there was considerable support for both sides throughout the state. In the end, up to 100,000 Kentuckians served for the Union side and up to 40,000 served on the Confederate side (Marshall 2010).

4

2.2

Aftermath of the Civil War

The literature on the costs of the war is substantial, and “costs” have been defined in di↵erent ways. Goldin and Lewis (1975) estimate the direct and indirect cost of the Civil War to the North and the South using aggregate data. Specifically, they estimate a consumption stream that would have occurred in the absence of the war and compare it to the stream that did occur. They find that the cost was very large for the entire country, but it was greater in the South. Social scientists have proposed a host of reasons why the South lagged behind the North by nearly any economic measure for a century after the conflict ended, beyond recovering from these losses. Scholars have highlighted low levels of human capital, relatively high fertility rates, over-reliance on cotton, and political institutions as factors that led to stalled economic development in the U.S. South (Wright 1986; Margo 1990; Naidu 2012; Sokolo↵ and Engerman 2000). Wright (1986) considers the reasons for stalled southern economic growth in the post-war era and argues that the “separateness” of the South is what led to its slower development. In particular, Wright (1986) posits that the South, so heavily and solely invested cotton agriculture, su↵ered upon the emancipation of slaves, which were financial assets to wealthy southerners.3 While there were other potential avenues of economic growth in the South, such as the mining of coal deposits or raising hogs, Wright (1986) argues that the South was so heavily invested in cotton agriculture that a switch to investment in other sectors was too costly for individuals and thus didn’t occur. Margo (1990) argues that poor white and black children in the South were undereducated (i.e. required to attend fewer days of school per year than richer white children) in an e↵ort by school boards to prevent children from seeking employment in the North upon reaching adulthood and instead maintain a large workforce in cotton agriculture. Lack of migration to the North resulted in the lack of convergence in real wages between the South and other sections of the country (Rosenbloom 1990). In general, the literature on the economic consequences of the Civil War emphasizes the experience of the southern U.S. Much less is written about the consequences of the war in border regions, where individual communities contributed troops to both sides. In border states, the “civil” nature of the Civil War is most apparent, and post-war social conflicts between people aligned with opposing sides may have had profound economic consequences. This study will contribute to the historical literature on the American Civil War by o↵ering new insight into the experience of border 3

There is a substantial literature on economic viability of slavery and whether slavery would have ended due to the lack of profitability in the South. See Conrad and Meyer (1958), Goldin (1976), and Fogel and Engerman (1974).

5

areas.

2.3

Outcomes of Union Army Veterans

There is a large literature on the post-Civil War outcomes of Union Army veterans. This literature typically does not focus on the consequences of military service, but rather uses Union army veterans as a representative sample of the northern male population for which high quality longitudinal data are available. Costa (1995, 1997) has used Union army veterans to study the impact of pension income on retirement and living arrangements; Eli (2015) has used this group to study income e↵ects on health; Salisbury (2014) uses Union Army widows to measure the e↵ect of a transfer with a marriage penalty on the rate of remarriage. One study that explicitly measures the impact of the war on veterans is Costa and Kahn (2008), which looks at how unit cohesion a↵ects later life outcomes, including migration behavior.

3

Data

Our dataset consists of linked military and census records. We describe the data from each source, and the procedure by which records are linked, sequentially.

3.1

Military Records

We begin with a collection of military records from the genealogical website fold3.com. The data we have collected are essentially indexes to compiled service records, which consist of muster rolls and other documents collected from the War Department and the Treasury Department. These records existed for both Union and Confederate soldiers; however, they are likely more complete for Union soldiers The indexes to these record collections contain the recruit’s regiment, full name, and (in some cases) age at enlistment. These indexes are extracted in their entirety for Kentucky, with 107,589 entries on the Union side and 50,304 entries on the Confederate side. Table 1 contains an illustration of the nature of the data extracted from these indexes. An obvious complication with using these indexes is that it is not clear when multiple entries refer to the same person. The first three entries in table 1 are men from the 3rd Union Cavalry named John Ewbanks, John Ubanks, and John Ebanks, respectively. The 4th entry is a man from the 55th Union Infantry, who is also named John Ewbanks. These names are all phonetic variants of one another, and could easily refer to the same person. Soldiers frequently re-enlisted in multiple

6

units, and if their names were spelled di↵erently on di↵erent muster rolls or were duplicated for some other reason, they could easily appear in this index multiple times. This poses a challenge for establishing the coverage of these records. In particular, estimating the coverage of these indexes will depend on assumptions that we make about which records are duplicates. In the top panel of table 1, we illustrate the least conservative grouping, in which we assume that phonetically identical names from the same regiment are the same person. In the example in table 1, this reduces the number of unique soldiers from 10 to 7. In the entire sample, this reduces the number of unique soldiers to 78,257 Union and 37,917 Confederate, for a total of 116,174 recruits from Kentucky (see table 2 for relevant statistics).4 Another possibility is to assume that all Union or Confederate soldiers with phonetically identical names are the same person, as illustrated in panel B of table 1; this reduces the number of soldiers in table 1 to 5, and it reduces the number of records in the complete sample to 64,309 (44,976 Union and 19,333 Confederate). How do these sample sizes compare with the likely number of military recruits from Kentucky? The best estimate (quoted in Marshall 2010) is that 90,000 to 100,000 Kentuckians enlisted on the Union side, while only 25,000 to 40,000 enlisted on the Confederate side. This suggests that our grouping in panel A of table 1 is likely to be the most accurate. It also suggests that men from Kentucky enlisted at a higher rate than the national average.5 The military indexes give us very little information other than the name of the recruit and the side on which he enlisted. Therefore, we need to match these indexes to other records in order to characterize these enlistees and their outcomes. A difficulty with using this data source is that the only information we can use to match military indexes to other records is first and last name. Although many enlistment records contain the recruit’s age, this is substantially more common in Union records: more than 80% of Union records contain the recruit’s age at enlistment, while only about 15% of Confederate records contain this information. Accordingly, we cannot use age at enlistment to match records without introducing severe systematic di↵erences in the accuracy of 4

These groupings are formed by creating NYIIS codes for both first and last names and grouping by these codes. When only first initials are given, they are grouped with full first names containing the same first initial. 5 Conventional figures for Civil War enlistment among whites are approximately 2,000,000 from the North and 800,000 from the South. In 1860, there were approximately 2.1 million while men ages 10-45 residing in the South and 6 million while men in the same age range residing in the North. These age ranges are based on data from the Union Army database (Fogel 2000), in which 99% of recruits were born between 1817 and 1847. The slightly expanded age range is intended to allow for measurement error in the reporting of age in the 1860 census. This implies that 33% of northern men in this age range enlisted, while 40% of southern men in this age range existed. There were approximately 275,000 white men in this age range residing in Kentucky in 1860. So, if enlistment patterns in Kentucky were similar to enlistment patterns elsewhere, this would imply that somewhere between 90,000 and 110,000 men from Kentucky enlisted in total, which is less than the actual enlistment of 115,000-140,000.

7

matches by Union or Confederate status. Furthermore, we cannot be sure how many individuals each unique name entry covers. Importantly, some names appear on both Union and Confederate rosters. To construct our list of names to match to census and death records, we group names by phonetic first and last name group, defined using NYIIS codes (Atack and Bateman 1992), and military side, i.e. Union or Confederate. We restrict to phonetic name groups that are uniquely identifiable as Union or Confederate, and we treat each phonetic group as a single individual. We also omit name groups that only include first initials, as we do not have sufficient information from these initials to accurately link our observations to other records.6 As an example, see panel C of table 1, in which only two of the five unique phonetic name groups listed would be included in our sample. As seen in table 2, this leaves us with 49,180 unique phonetic name groups to match, 38,318 of which are Union and 10,862 of which are Confederate.

3.2

Matches to 1860 census

We match our sample of unique Union and Confederate names to the census of 1860 using records available from ancestry.com via the NBER. Again, our challenge is that the only linkable information we have from military data is the soldier?s name. So, to facilitate matching to the census, we impose certain restrictions on our target sample of census records. First, because our sample of recruits comes from regiments of white males, we limit our search to white males in the census. A sample of Union Army veterans indicates that 99% of Union recruits were born between 1817 and 1847 (Fogel 2000). Assuming a similar age range in the Confederate army, and allowing for some error in the reporting of ages, we further restrict our search of the 1860 census to men born between 1815 and 1850. Finally, we restrict the geographic area in which we search for these soldiers. We impose these restrictions on our target sample in order to minimize error in matching. An unrestricted match to the 1860 census based on name alone would yield many potential matches, most of which would be incorrect. Using information about the prior probability that recruits have other characteristics can improve the accuracy of our matches. Take, for example, residential location. Given that our recruits enlisted in Kentucky regiments, it is overwhelmingly likely that they resided in Kentucky at the time of enlistment, which occurred between 1861 and 1865. Companies 6

This restriction reduces the number of Confederate recruits relative to Union recruits, since almost 20% of Confederate records list only a first initial, while very few Union recruits list only a first initial. This can be seen in table 2. We find that, in our Confederate sample of names, the regiment that the soldier enlisted in explains about 12% of the variation in whether or not a full first name is reported, and we do not find evidence that the socioeconomic status of the soldier’s surname is related to the probability of reporting a full first name. As such, we believe that reporting only a first initial reflects record keeping practices of individual regiments or companies rather than systematic socioeconomic di↵erences.

8

were typically organized locally, and regiments were named after the state that enlistees were from. So, we believe it is fair to assume that recruits were more likely to reside in Kentucky in 1860 than elsewhere; as such, matches residing in Kentucky are more likely to be correct than matches residing elsewhere. We perform two versions of our matching procedure: one in which we match military records to white men ages 10-45 residing in Kentucky (275,999 records in target sample), and one in which we match our military records to white men ages 10-45 residing in states surrounding Kentucky (3,610,482 records in target sample).7 We match names by searching for exact phonetic first name and surname matches between the military records and the target census sample, then by comparing the similarity of the first and last names using the jaro-winkler algorithm (Ruggles et al 2010). We discard matches with a string similarity score of less than 0.9. Table 3 contains information on matching rates using both approaches. Not surprisingly, matching to an expanded geographic area increases the fraction of military records matched to at least one census record, from around 43% to 66%. However, it decreases the fraction of records that are matched uniquely to the target sample, from around 25% to 18%. Moreover, it appears that the matches made exclusively to Kentucky are more accurate. Recall that we have information on ages for most of the Union recruits in our sample. While we do not perform matches to the census using this information, we can use it to check the accuracy of our results. Specifically, for individuals with an age of enlistment recorded on their military record, we can estimate Agemil =

0

+

1 Age1860

+u

If a match is correct, the age on the military record (Agemil ) should be essentially the same as the age in the census record (Age1860 ). So, a sample of correct matches should yield an estimated intercept close to zero and a slope close to one. In the bottom panel of table 3, we estimate this regression equation under two specifications: (i) using only records that uniquely matched to the 1860 census; and (ii) using all matched records, weighting multiple matches by 1/N , where N is the number of census records that match the military record in question. We estimate these specifications for three samples: (i) a sample matched to all states surrounding Kentucky; (ii) a sample matched to Kentucky only; (iii) and a sample matched to Missouri only, as a sort of placebo test. The first four columns of panel B of table 3 indicate that using unique matches between military 7 These states are: Kentucky, Tennessee, Missouri, Illinois, Indiana, Ohio, Virginia, Arkansas, Mississippi, Alabama, Georgia, North Carolina, and South Carolina.

9

records and the 1860 census introduces less error than using weighted multiple matches. And, these results indicate that matching to Kentucky only is more accurate than matching to all states bordering Kentucky. Restricting the target sample to Kentucky will cause us to miss (or erroneously match) recruits who migrated to Kentucky subsequent to 1860. However, it appears that expanding the target sample introduces enough false matches that we are better o↵ with the restriction. The last two columns of the table indicate that matches to Missouri alone are extremely inaccurate, which gives us further confidence that our matches to Kentucky are of a high quality.8

3.3

Matches to 1880 census

We match our recruits from the 1860 census to the 1880 100% census sample (NAPP). Here, we make use of the demographic information we obtain from the 1860 census when matching our records. We search the entire 1880 census for records that exactly match our 1860 census records on the following dimensions: birth place, phonetic first and last name codes, sex, and race. We restrict birth year in 1880 census to be no more than three years before or after birth year in the 1860 census. Finally, we discard matches in which the index measuring the similarity of names across census records (using the jaro-winkler algorithm) is less than 0.9. These procedures approximately follow Ruggles et al (2010). Using this procedure, we are able to uniquely match 30% of our Union soldiers and 29% of our Confederate soldiers. This match rate compares favorably to other studies that perform automated record linkages (Ferrie 1996; Ruggles et al 2010; Abramitzky et al 2012).

4

Empirical Approach

We are interested in understanding how service in the military a↵ected locational choices after the Civil War. We typically model migration as an investment, which maximizes a person’s expected lifetime earnings. Importantly, we usually think of migration as welfare maximizing: if people migrate in response to regional wage di↵erentials, they are e↵ectively sorting themselves into regions where labour is relatively productive. Moreover, if people migrate to regions that complement their individual skill profiles, as much of the existing research on migrant selection contends, they are 8

This “check” on the accuracy of our matches is necessarily driven by Union recruits, as they comprise the overwhelming majority of records with age information. However, we have no reason to believe that Confederate recruits were less likely to come from Kentucky than Union recruits. When we match to all states surrounding Kentucky, we end up finding a greater fraction of Confederate matches in Kentucky than Union matches: 30% of our Confederate matches reside in Kentucky, whereas 26.5% of our Union matches live in Kentucky. So, we are confident that restricting our target sample to Kentucky improves match accuracy overall.

10

sorting into the regions in which they are individually most productive. In the case of civil conflict, migration may occur for other reasons. In particular, animosity among combatants may generate migrations which would never have happened for economic reasons in the absence of the conflict. As such, this represents an additional “cost” imposed by civil war. Our aim with this paper is to measure the degree to which conflict among recruits from opposing sides generated a migration response after the Civil War. To cast this in economic terms, we can think of a person i’s earnings in county c (Yic ) depending on both individual ability (Aic ) and “social capital” (Sic ), both of which are person and location specific. A person will choose to locate in the county that maximizes Y (Aic , Sic ) net of migration costs; if this is the person’s home county, he will choose not to migrate. Our hypothesis is that the Civil War a↵ected Sic . In particular, in counties more sympathetic to the North, Sic should have fallen for Confederate recruits and risen for Union recruits; in counties more sympathetic to the South, Sic should have fallen for Union recruits and risen for Confederate recruits. This will a↵ect observed migration behavior in two key ways: (1) Recruits who have experienced a reduction in Sic in their home county should be more likely to migrate; (2) Migrant recruits should be more likely to select a destination in which Sic has increased for recruits from their side. We use two county-level indicators to infer relative “social capital” to Union and Confederate recruits after the war: (1) the county’s share of the presidential vote going to Stephen A. Douglas – the Democratic candidate – in the 1860 election; (2) the fraction of the county’s population that was enslaved in 1860. As we will show in the next section, both are strong predictors of military side. Conditional on other characteristics, a 10 percentage point increase in vote share to Douglas generates a 6 percentage point increase in the probability of serving in the Confederate army, and a 10 percentage point increase in the fraction enslaved generates a 6.5 percentage point increase in the probability of joining the Confederate army; this is significant at the 1% level. So, we will use both of these indicators to proxy for a county’s degree of social alignment with the South. We then test whether the propensity to leave a county depends di↵erently on these indicators for Union and Confederate recruits. We also test whether migrants from Union and Confederate sides sorted di↵erentially into places more sympathetic to the South, measured as Douglas vote share or fraction enslaved in 1860 for intrastate migrants, and region of residence in 1880 for interstate migrants. To determine whether Union recruits were more likely to leave more “Confederate” counties,

11

we estimate the following equation by OLS: Mij,1880 = ↵ +

1 Uij

+

2 Sj,1860

+

3 Uij

⇥ Sj,1860 + Xi,1860 +

j

+ uij

(1)

Here, Mij is an indicator equal to one if person i from county j had migrated by 1880; Uij is equal to 1 if this person served in the Union army; Sj,1860 is an indicator of sympathy for the Confederacy in county j in 1860; Xi,1860 is a matrix of individual characteristics observed in 1860, including age and birthplace fixed e↵ects;

j

is an 1860 county fixed e↵ect. We expect to find

3

> 0.

A complication with this approach is that Union and Confederate recruits are drawn from systematically di↵erent parts of the skill distribution: Confederate recruits are more skilled on average. So, we may find that Union recruits are more likely to leave “Confederate” counties if these counties are more complementary to skilled individuals. In other words, di↵erences in migration propensities may work through di↵erences in individual skill and not di↵erences in social capital. We show that migrants from counties more sympathetic to the Confederacy are not negatively selected on 1860 socioeconomic status, and that our results are robust to controlling for interactions between ex ante socioeconomic status and our indicators of social alignment with the South. While Union and Confederate recruits may also be selected on unobservable skill, it is unlikely that counties more aligned with the South will systematically favor workers with unobservable skills but not observable ones. Thus, we argue that it is unlikely that di↵erences in skill between Union and Confederate recruits can explain these results. To determine whether Union recruits were more likely to sort into less “Confederate” counties, we estimate the following, using a sample of internal migrants within Kentucky: Sik,1860 = ↵ + Uijk + Xi,1860 +

j

+ uij

(2)

Here, Sik,1860 is an indicator of social alignment with the South in county k, where person i is residing in 1880; and Uijk is an indicator equal to one if person i who migrated from county j to county k between 1860 and 1880 served in the Union army. The remaining variables are defined as above. Here, we expect to find

< 0. We also estimate regressions with indicators for residence in

each region in 1880, using a sample of interstate migrants. We predict that enlistees on the Union side should be more likely to move north than enlistees on the Confederate side. Again, this is complicated by systematic di↵erences in skill between Union and Confederate recruits. Because we are able to control for ex ante occupational attainment and family wealth, we 12

can rule out the hypothesis that di↵erences in locational choices are entirely driven by observable skill. One specific concern is that Union veterans were eligible to acquire land under the Homestead Act of 1862, while Confederate veterans were excluded until 1867. So, Union veterans may have disproportionately migrated to areas with better land, since they had the first opportunity to do so. To argue that this does not explain our regional location results, we control for 1880 farm value per acre and the rate of farm ownership, which are the characteristics that are available at the county level in 1880. If we still estimate a significant

, this means that regional locational di↵erences

cannot be explained by regional di↵erences in land quality. It is likely that social and individual capital interact in determining a person’s earnings. For instance, occupations higher in the skill distribution – such as managers and officials – may benefit more from social capital that occupations that are lower in the skill distribution – like common laborers. Thus, Union and Confederate migrants from di↵erent counties may be di↵erently selected in terms of occupational status or wealth. We test this explicitly by estimating the following regressions separately for Union and Confederate veterans: Mij,1880 = ↵ + Yi,1880 = ↵ +

1 Yi,1860

+

1 Mij,1880

2 Yi,1860

+

⇥ Sj,1860 + Xi,1860 +

2 Mij,1880

⇥ Sj,1860 +

3 Yi,1860

j

+ uij

(3)

+ Xi,1860 +

(4)

j

Yi denotes log occupational earnings, which is the only available measure of skill in both 1860 and 1880. In equation (3), the parameter higher S. In particular, if

2

2

captures the selectivity of migrants from counties with

> 0, this means that migrants from counties with larger S are more

positively selected than migrants from counties with lower S. If

2,U nion

>

2,Conf ed. ,

this means

that Union migrants from counties more aligned with the Confederacy are more positively selected than Confederate migrants from these counties. This is what we would expect if S a↵ects the earnings of skilled individuals more than the earnings of unskilled individuals. In equation (4),

2

captures a di↵erential return to migration for migrants from counties with higher S. If Union and Confederate recruits are di↵erently selected out of Confederate counties on unobservable skills, we should expect to see a larger return to migration for Union recruits from counties with large S. Lastly, we are concerned that our results may be driven by pre-existing social divisions, and not by the experience of military service itself. If this is the case, then we should see similar di↵erences in migration behavior among the younger brothers of Union and Confederate recruits from Kentucky. We explicitly test this with our linked data by estimating our main specifications

13

on a sample of younger brothers.

5

Results

We present preliminary results using our sample of soldiers’ names that are matched uniquely to Kentucky in 1860 (12,440 individuals). We are able to uniquely match 3,553 of these men to the census of 1880. We use only these samples in the results that follow.

5.1

Characteristics of Soldiers in 1860

Figure 2 illustrates the geographic distribution of Union and Confederate soldiers within Kentucky. Darker counties contain more enlistees, as a percentage of the total number of enlistees on each side. The distribution of soldiers on both sides is relatively dispersed: the largest share to be contained in a single county is 6% for Union soldiers and 4% for Confederate soldiers. Still, there are certain systematic di↵erences in the location of soldiers who ultimately enlist on each side. There is a concentration of Union recruits in coal-producing areas of the state, specifically in the lower portion of the eastern mountains and coalfields, and in the western coalfields. There are concentrations of Confederate enlistees from the northeastern agricultural (“Bluegrass”) region and around the Mississippi Plateau in the southwest portion of the state, which is also an agricultural region. There is also a concentration of Confederates in the eastern part of the state along the border with Virginia. It is, however, notable that there are many overlapping or adjacent areas of Union and Confederate concentration. In table 4, we compare average characteristics of Union and Confederate soldiers. The first column contains mean values of each variable for Union soldiers, the second columns contains means for Confederates, and the third column contains means for all white men in Kentucky between the ages of 10 and 45. The fourth column contains results from an OLS regression of an indicator for Union status on all characteristics together. As a group, soldiers were younger and less likely to be married than the general population, which is not surprising. They were also more likely to be native to Kentucky or native to the United States. Comparing Union and Confederate soldiers, a number of di↵erences are apparent. On average, Union soldiers were older, more likely to be married, and less likely to live with a parent. Figure 3 additionally shows the age distribution for soldiers from both sides. There are more Union soldiers in their 30s and 40s in 1860; however, there are also more Union soldiers in their early teens. This

14

likely indicates that enlistment near the end of the war was more skewed toward the Union side than enlistment at the beginning of the war By the end of the war, both armies were enlisting soldiers who were much older and younger than were enlistees at the beginning of the war, when most soldiers were in their early 20s. Table 4 also indicates large di↵erences in nativity. Confederate enlistees were much more likely to be born in Kentucky or in the South generally (including Kentucky). Union soldiers were much more likely to be born in the Northeast, Midwest, or abroad. Evidence also points to di↵erential selection of Confederate soldiers on socioeconomic characteristics. Confederate soldiers systematically came from counties with more slaves, greater value of property per family, and more people employed in agriculture. While we do not have data on the individual wealth of everyone in our sample, we find that Confederate soldiers typically had surnames that were associated with greater value of real estate and more white collar employment in 1850. These findings are consistent with men who had greater ties to slavery being more likely to join the Confederate army. We also find significant di↵erences in voting patterns. Men who joined the Confederate army tended to live in counties with a greater vote share going to the Democratic Party in the 1860 presidential election; Conversely, Union soldiers came from counties more likely to vote for John Bell, an alternative candidate from the Constitutional Union party who opposed the westward expansion of slavery.

5.2

Outcomes for Soldiers in 1880

Table 5 contains additional summary statistics for our sample of soldiers who are matched to the 1880 census. This table includes average outcomes in 1880, as well as average 1860 characteristics that we have only collected for this sample. These are largely consistent with table 4, in that they point to Confederate recruits being of higher socioeconomic status ex ante. The table also points to systematic di↵erences in locational outcomes for Union and Confederate soldiers. We discuss these di↵erences in detail below. 5.2.1

Migration Propensity

From table 5, we can see that roughly 70% of our sample still resided in Kentucky as of 1880; however, approximately 50% of our sample had moved between counties by 1880. This is true of both Union and Confederate veterans. In table 6, we estimate OLS regressions of an indicator for having moved between counties (columns 1-4) or states (columns 5-8) on an indicator for having served in the Union army, as well as other individual characteristics and 1860 county fixed 15

e↵ects. We find no evidence that Union recruits were more likely to migrate than their Confederate counterparts. We find some evidence that migrants were positively selected in terms of occupational attainment but negatively selected in terms of family wealth. As farmers typically owned a large number of assets but had a relatively low occupational income score, this likely indicates that landowning farmers were unlikely to move, relative to semi skilled or skilled workers. In table 7, we estimate the impact of home county vote share to Douglas and the home county incidence of slavery on the propensity to migrate among Union and Confederate recruits. We find that Union recruits are significantly more likely that Confederate recruits to leave counties with a larger vote share going to Douglas in the 1860 presidential election. To address concerns that this is driven by county-level di↵erences in the return to skill, we include indicators for occupational class and interactions between these indicators and Douglas vote share in column (2). This has very little impact on our results; moreover, we do not find evidence that migrants from counties with a greater Douglas vote share are negatively selected on skill, which would be a necessary condition for skill di↵erences among Union and Confederate recruits to explain our results. In column (3), we proxy observable skill with log family wealth in 1860, and we find similar results. In columns (4)-(6), we use the 1860 fraction enslaved to measure social alignment with the South, and we find broadly similar results. 5.2.2

Migration Destination

In figure 4, we map the locations of interstate migrants from the Union and Confederate army in 1880. There appear to be clear locational di↵erences: Union recruits are more likely to move north, and Confederate recruits are more likely to move south and west. In table 8, we estimate di↵erences in locational choices of interstate migrants by military side. We regression indicators for having moved to the South, Northeast, Midwest and West on an indicator for having served in the Union army, as well as other 1860 characteristics including age, birthplace and county of residence fixed e↵ects. We find that Union recruits were significantly more likely to migrate to the Midwest, and less likely to migrate to the South or the West. This is conditional on ex ante occupational attainment, literacy, and family wealth, so di↵erences in destination regions cannot be explained by di↵erences in observable skill. This is consistent with recruits moving to areas of the country where the social returns to their military service are highest.9 9

As the West was sparsely populated, it migration to the West should have been associated with a smaller “penalty” for serving on the Confederate side.

16

As mentioned above, a concern is that these results are not driven by regional di↵erences in the social returns to military service, but by the Homestead Act of 1862. Because Confederate veterans were excluded until 1867, they may have been excluded from acquiring the best available land, if this land was claimed first. If the best land was in the Midwest, this might generate our results. To address this concern, we control for two salient 1880 county characteristics: average farm value per acre, and the ownership rate in agriculture. If di↵erences in destination region are totally explained by the fact that Union recruits could access better quality land, then including these controls should wipe out any systematic regional di↵erences in location by military side. While this does substantially reduce the e↵ect of being in the Union army on the probability of living in the South in 1880, it does not substantially alter the e↵ect on the probability of living in the Midwest or West. In table 9, we estimate di↵erences in locational choices of intrastate migrants in Kentucky. We find that intrastate migrants from the Union side migrated to counties with a smaller Douglas vote share in 1860, and with a lower incidence of slavery in 1860. Again, this is is robust to controlling for ex ante observable skill. 5.2.3

Migrant Selection

In tables 10 and 11, we explore di↵erences in the selection of migrants from the Union and Confederate side. We believe that it is likely that social and individual capital are complementary: strong social networks have a larger e↵ect on the earnings of skilled workers than unskilled workers. Thus, we should expect to see social alignment with the Confederacy a↵ecting the migration decisions of skilled and unskilled workers di↵erently. In table 10, we look at di↵erences in the selection of migrants overall. We regress an indicator for having migrated between 1860 and 1880 on measures of observable skill in 1860 separately for Union and Confederate recruits, and we test whether or not our coefficients of interest are significantly di↵erent for Union and Confederate recruits. We find some evidence that migrants from the Union side were more positively selected than movers from the Confederate side. In particular, being from the top two tiers of the skill distribution (white collar or skilled blue collar) strongly predicts migration among Union veterans; however, this is not the case for Confederate veterans. The di↵erence in the e↵ect of white collar status on migration between samples is significant at the 5% level. This is consistent with what we believe about the complementarity between social capital and skill. As a rule, Kentucky became less sympathetic to the North and more sympathetic to the South after the Civil War. So, we should expect Union 17

“leavers” to be disproportionately more skilled if the social penalty they faced after the war mostly a↵ected skilled workers. Importantly, this finding remains if we classify “migrants” as interstate migrants only. In table 11, we estimate equations (3) and (4). In the top panel, we look at whether or not Union recruits from more “Confederate” counties were more positively selected than Confederate recruits from “Confederate” counties, in terms of ex ante observable skill and unobservable skill (measured as occupational attainment in 1880, conditional on occupational attainment in 1860). We find evidence that this is the case. We also find some evidence that the return to migration was greater for Union recruits than Confederate recruits. In the bottom panel, we look at di↵erences in the selection of migrants from counties with a higher incidence of slavery. Here, we find that Union migrants from counties with more prevalent slavery are more positively selected than Confederate migrants in terms of family wealth, but not occupational attainment. Similarly, we do not find that Union migrants from counties with more slaves attained a higher occupational status in 1880 than Confederate migrants from these counties. However, given the close association between slavery and agriculture, we are a bit concerned about how the low placement of farmers in our occupational ranking is a↵ecting these estimates. We should also note that Union recruits may have been more positively selected in predominantly Confederate counties: perhaps skilled men were more likely to “go against the grain.” We test whether or not Union recruits were more skilled ex ante (measured by occupational attainment and family wealth) when they came from counties more aligned with the Confederacy. We do not find any evidence that this is the case. However, we will grant that they may have been selected on unobservables, which may contribute to their di↵erent occupational outcomes in 1880. 5.2.4

Brothers of Civil War Recruits

We ave argued that social penalties associated with serving on the Union or Confederate side generated a migration response after the Civil War. However, it is possible that these social divisions existed before the war, and we would have observed the same migration behavior in the absence of the conflict. To test this conjecture – and the conjecture that recruits were punished or rewarded for “ideology” rather than military service explicitly – we estimate our baseline specifications using a sample of younger brothers of Union and Confederate veterans. If men who did not fight in the war (but are from families that sided with one army or the other) behave similarly to combatants, this suggests that the experience of actually serving on one side or the other is not so important. 18

We present these results in table 12. While some of the patterns we observe for recruits hold true for their younger brothers, many others do not. This suggests to us that military service on one side or the other was an important determinant of migration behavior.

6

Conclusion

Our results suggest that social divisions among Union and Confederate sides following the Civil War had large e↵ects on migration behavior. We measure a county’s social alignment with the Confederacy – which increases the relative return to having served in the Confederate army – by the vote share the Stephen A. Douglas in the 1860 presidential election, and the fraction of the population that was enslaved in 1860. We document systematically di↵erent migration behavior among Union and Confederate veterans, which depends on the ex ante social alignment of their county of origin with the Confederacy. Our results suggest that recruits migrated as a consequence of social divisions created by the war.

References [1] Abramitzky, Ran, Leah Platt Boustan, and Katherine Eriksson (2012). “Europe’s Tired, Poor, Huddled Masses: Self-Selection and Economic Outcomes in the Age of Mass Migration.” American Economic Review. 102(5): 1832-1856. [2] Annan, Jeannie and Christopher Blattman (2010). “The Consequences of Child Soldiering.” Review of Economics and Statistics. 92(4): 882-898. [3] Atack, Jeremay and Fred Bateman (1992). “Matchmaker, Matchmaker, Make Me a Match: A General Personal Computer-Based Matching Program for Historical Research” Historical Methods. 25(2):53-65. [4] Beard, Charles A. The Rise of American Civilization. New York: The Macmillan Company, 1927. [5] Blattman, Christopher, and Edward Miguel. “Civil War.” Journal of Economic Literature, 48(1): 3- 57, 2010. [6] Bleakley, Hoyt, Louis Cain, and Joseph Ferrie. “Amidst Poverty and Prejudice: Black and Irish Civil War Veterans,” in Institutions, Innovation, and Industrialization: Essays in Economic History and Development, Avner Greif, Lynne Keisling, John V.C. Nye (eds.), 2014. (Festschrift volume for Joel Mokyr.) [7] Bundervoet, Tom, Philip Verwimp, and Richard Akresh (2006). “Health and Civil War in Rural Burundi.” Journal of Human Resources. 44(2): 536-563.

19

[8] Cain, Louis P. and Sok Chul Hong. “Survival in 19th Century Cities: The Larger the City, the Smaller Your Chances,” Explorations in Economic History, October 2009. [9] Cohran, Thomas C. “Did the Civil War Retard Industrialization?” Mississippi Valley Historical Review, XLVIII (September 1961), 197-210. [10] Collins, William J. and Marianne H. Wanamaker (2014). “Selection and Economic Gains in the Great Migration of African Americans: New Evidence from Linked Census Data.” American Economic Journal: Applied Economics. 6(1): 220-252. [11] Clubb, Jerome M., William H. Flanigan, and Nancy H. Zingale. Electoral Data for Counties in the United States: Presidential and Congressional Races, 1840-1972. ICPSR08611-v1. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 200611-13. doi:10.3886/ICPSR08611.v1 [12] Conrad, Alfred H. and John R. Meyer. “The Economics of Slavery in the Ante Bellum South.” The Journal of Political Economy, 66(2): 95-130. April 1958. [13] Costa, Dora L. “Pensions and Retirement: Evidence from Union Army Veterans.” Quarterly Journal of Economics. May 1995, 110(2): 297-320. [14] Costa, Dora L. “Displacing the Family: Union Army Pensions and Elderly Living Arrangements.” Journal of Political Economy. December 1997, 105(6): 1269-1292. [15] Costa, Dora L, and Matthew Kahn. Heroes and Cowards: The Social Face of War. 2008. Princeton University Press. [16] Eli, Shari J. (2015). “Income E↵ects on Health: Evidence from Union Army Pensions.” Journal of Economic History. 75(2): 448-478. [17] Engerman, Stanley. “Slavery as an Obstacle to Economic Growth in the United States: A Panel Discussion,” with Alfred H. Conrad and others. Chapter 2 in Paul Finkelman, ed., Articles on American Slavery, Volume 10, Economics, Industrialization, Urbanization and Slavery, New York: Garland Press, 1989: 28-70. [18] Ferrie, Joseph P. (1996). “A New Sample of Americans Linked from the 1850 Public Use Micro Sample of the Federal Census of Population to the 1860 Federal Census Manuscript Schedules.” Historical Methods. 29: 141- 156. [19] Fogel, Robert W. (2000). Public Use Tape on the Aging of Veterans of the Union Army: Military, Pension, and Medical Records, 1860-1940, Version M-5. Center for Population Economics, University of Chicago Graduate School of Business, and Department of Economics, Brigham Young University. [20] Fogel, Robert and Stanley Engerman. Time on the Cross: The Economics of American Negro Slavery. New York: W.W. Norton and Company. 1974. [21] Goldin, Claudia and Frank Lewis. “The Economic Cost of the American Civil War: Estimates and Implications,” Journal of Economic History. June 1975 35: 299-326. [22] Goldin, Claudia and Frank Lewis. “The Post-Bellum Recovery of the South and the Cost of the Civil War: A Comment,” Journal of Economic History. June 1978 38: 487-92.

20

[23] Goldin, Claudia. Urban Slavery in the American South, 1820 to 1860: A Quantitative History. Chicago, IL: University of Chicago Press. [24] Haines, Michael R., and Inter-university Consortium for Political and Social Research (2010). Historical, Demographic, Economic, and Social Data: The United States, 1790-2002 [Computer file]. ICPSR02896-v3. Ann Arbor, MI: Inter-university Consortium for Political and Social Research [distributor], 2010-05-21. doi:10.3886/ICPSR02896 [25] Harrison, Lowell H. (1975). The Civil War in Kentucky. Lexington: The University Press of Kentucky. [26] Margo, Robert. Race and Schooling in the South, 1880-1950: An Economic History, NBER Monograph Series on Long-Term Factors in Economic Development. Chicago: University of Chicago Press, 1990. Pp. vii, 164. [27] Marshall, Ann. Creating a Confederate Kentucky: The Lost Cause and Civil War Memory in a Border State. Chapel Hill: University of North Carolina Press, 2010. [28] Miguel, Edward and Gerard Roland (2011). “The Long-Run Impact of Bombing in Vietnam.” Journal of Development Economics. 96(1): 1-15. [29] Naidu, Suresh. “Su↵rage, Schooling, and Sorting in the Post-Bellum U.S. South.” Working Paper, 2012. [30] Rosenbloom, Joshua. “One Market or Many? Labor Market Integration in the Late NineteenthCentury United States.” Journal of Economic History, 50 (March 1990): 85-108 [31] Ruggles, Steven J., Trent Alexander, Katie Genadek, Ronald Goeken, Matthew B. Schroeder, and Matthew Sobek. Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database]. Minneapolis: University of Minnesota, 2010. [32] Salisbury, Laura. “Women’s Income and Marriage Markets in the United States: Evidence from Civil War Pensions,” NBER Working Paper No. 20201, June 2014. [33] Salsbury, Stephen. “The E↵ect of the Civil War on American Industrial Development,” in Ralph Andreano, ed, The Economic Impact of the American Civil War. New York: Schenkman Publishing Co., 1967. [34] Sokolo↵, Kenneth and Stanley Engerman. “Institutions, Factor Endowments, and Paths of Development in the New World.” Journal of Economic Perspectives, 14(3): 217-232, Summer 2000. [35] Wright, Gavin. Old South, New South: Revolutions in the Southern Economy Since the Civil War, Baton Rouge: Louisiana State Press, 1986.

21

7

Tables and Figures Table 1: Military Data: Example Side

Regiment

Name

Panel A: phonetic name + regiment groups Union 3rd Cavalry John Ewbanks Union 3rd Cavalry John Ubanks Union 3rd Cavalry John Ebanks Union 55th Infantry John Ewbanks Union 1st Cavalry Jefferson Eubanks Confederate Kirkpatrick's Battalion John J Ewbank Confederate Kirkpatrick's Battalion J J Eubank Confederate 10th Infantry Napolean Ewbanks Confederate 12th Cavalry Napolean Eubanks Confederate 19th Infantry F Eubanks Panel B: phonetic name + union/confederate groups Union 3rd Cavalry John Ewbanks Union 3rd Cavalry John Ubanks Union 3rd Cavalry John Ebanks Union 55th Infantry John Ewbanks Union 1st Cavalry Jefferson Eubanks Confederate Kirkpatrick's Battalion John J Ewbank Confederate Kirkpatrick's Battalion J J Eubank Confederate 10th Infantry Napolean Ewbanks Confederate 12th Cavalry Napolean Eubanks Confederate 19th Infantry F Eubanks Panel C: names included in final sample Union 3rd Cavalry John Ewbanks Union 3rd Cavalry John Ubanks Union 3rd Cavalry John Ebanks Union 55th Infantry John Ewbanks Union 1st Cavalry Jefferson Eubanks Confederate Kirkpatrick's Battalion John J Ewbank Confederate Kirkpatrick's Battalion J J Eubank Confederate 10th Infantry Napolean Ewbanks Confederate 12th Cavalry Napolean Eubanks Confederate 19th Infantry F Eubanks

22

Table 2: Military Data: Statistics

Total # Records % of total # unique phonetic last name + phonetic first name + regiment groups # unique phonetic last + phonetic first + union/confederate groups # groups with only first initials available % of groups with only first initials available # unique phonetic last + phonetic first + union/confederate groups, excl groups with only first initials, identifiable as union/confederate only % of total Note: about 75% of phonetic name groups contain unique entries.

23

Union

Confederate

Total

107,589 68.1% 78,257 44,976 129 0.3%

50,304 31.9% 37,917 19,333 3,578 18.5%

157,893 100% 116,174 64,309 3,707 5.8%

38,318

10,862

49,180

77.9%

22.1%

100%

Table 3: Matches between Military Data and 1860 Census: Statistics Panel A. Linkage Rates to Census by Target Sample Matched to all states near KY Union Confederate

24

# military records matched at least once to census Mean # matches per military record (of matched) # unique matches to census % of military records matched to at least one census Conventional match rate % of matches living in Kentukcy

25,461 11.86 6,503 66.4% 17.0% 26.5%

7,103 9.00 2,008 65.4% 18.5% 29.5%

Matched to KY Union Confederate *** ** *** ***

16,472 2.12 9,529 43.0% 24.9% -

4,588 1.81 2,911 42.2% 26.8% -

*** *** -

Panel B. Test of Accuracy of Matched Records Sample: Coefficient on Military Age Constant R-squared Observations

Matched to all states near KY Unique Weighted 0.501 10.3 0.185 5,944

0.214 18.0 0.027 302,022

Matched to KY Unique Weighted 0.632 7.0 0.294 9,391

0.474 11.4 0.135 35,393

Matched to MO (placebo) Unique Weighted 0.051 22.3 0.001 4,323

0.048 22.5 0.001 28,299

Table 4: Characteristics of Union and Confederate Soldiers, 1860 Mean Comparison

OLS Regresion All men 10-45 in Kentucky, 1860

Union

Confederate

Age

22.398

22.031

**

23.895

Married

0.319

0.270

***

0.370

Household head

0.302

0.251

***

0.371

Lives with parent

0.442

0.481

***

0.385

Born Kentucky

0.749

0.820

***

0.722

Born south (incl. Kentucky)

0.863

0.933

***

0.834

Born northeast

0.018

0.010

***

0.021

Born midwest

0.043

0.023

***

0.038

Immigrant

0.077

0.034

***

0.106

County % agricultural

0.754

0.788

***

0.766

County % urban

0.090

0.081

County % slave

0.148

0.189

County % free black

0.008

0.009

Property per family ($1000)

3.873

4.909

County ag value per acre

18.524

22.072

22.403

County mean farm size

2.371

2.315

2.210

County churches per 100 people

0.193

0.187

0.187

County value per church

2.233

2.234

2.675

Vote share: Bell

0.454

0.419

***

0.451

Vote share: Douglas

0.357

0.441

***

0.353

Vote share: Breckenridge

0.178

0.134

***

0.186

Presidental voter turnout

0.667

0.699

***

0.671

Surname: mean prop., 1850 ($1000)

1.240

1.739

***

1.503

Surname: % white collar, 1850

0.057

0.066

***

0.065

Surname: % farmer, 1850

0.627

0.625

0.623

Surname: % laborer, 1850

0.064

0.061

0.062

0.117 ***

0.176 0.009

***

4.576

Constant Observations R-squared

Dependent variable = 1 if Union -0.002** (0.001) 0.019 (0.015) 0.020 (0.019) -0.013 (0.011)

0.100*** (0.026) 0.081*** (0.026) 0.153*** (0.021) 0.041 (0.088) -0.254** (0.112) -0.650*** (0.226) 1.229 (1.765) 0.002 (0.012) -0.001 (0.002) 0.010 (0.019) 0.249* (0.143) 0.004 (0.008) -0.597*** (0.101) -0.189 (0.133) -0.161 (0.182) -0.002*** (0.000) -0.058 (0.053) -0.002 (0.025) 1.204*** (0.161)

9,529

2,911

275,999

10,346 0.094

Note: Stars next to mean comparison refer to significance of coefficient on 1860 characteristic in a univariate regression of union status on that characteristic. For county-level characteristics, standard errors are clustered at county level. Regression in final column also clusters standard errors by county.

25

Table 5: Summary Statistics of Linked 1860-1880 Census Data Union 1880 Characteristics Age Married # Children in household Lives in Kentucky Lives elsewhere in south Lives in northeast Lives in midwest Lives in west White collar Semi-skilled Farmer Laborer No occupation 1860 Characteristics Family wealth ($1000) Household head Married Literate Parent white collar Parent semil-skilled Parent farmer Parent laborer Parent no occupation White collar Semi-skilled Farmer Laborer No occupation

Mean Confederate

42.136 0.886 3.490 0.697 0.061 0.022 0.210 0.010 0.081 0.109 0.647 0.137 0.026

41.951 0.864 3.207 0.719 0.093 0.010 0.157 0.021 0.121 0.100 0.654 0.096 0.028

2.109 0.309 0.320 0.901 0.049 0.091 0.710 0.052 0.098 0.045 0.117 0.409 0.251 0.179

6.074 0.279 0.289 0.935 0.088 0.078 0.733 0.040 0.060 0.068 0.092 0.448 0.225 0.167

26

Sample size Union Confederate * *** *** ** *** ** *** *** *** * *** ***

** *

2,743 2,743 2,743 2,743 2,743 2,743 2,743 2,743 2,743 2,743 2,743 2,743 2,743

810 810 810 810 810 810 810 810 810 810 810 810 810

2,716 2,743 2,743 2,726 1,260 1,260 1,260 1,260 1,260 1,483 1,483 1,483 1,483 1,483

798 810 810 797 397 397 397 397 397 413 413 413 413 413

Table 6: Determinants of Migration between 1860 and 1880 (1) (2) (3) (4) Moved counties between 1860 & 1880

Dependent variable Union

0.005 (0.020)

0.014 (0.021)

0.023 (0.026) -0.054*** (0.013) 0.024 (0.033) 0.096*** (0.029)

Log family wealth, 1860 Literate, 1860 Log occupational score, 1860

-0.011 (0.021) -0.048*** (0.010) -0.003 (0.031)

White collar, 1860

(6) (7) (8) Moved states between 1860 & 1880

-0.005 (0.017)

-0.005 (0.018)

0.013 (0.023) -0.003 (0.012) 0.029 (0.029) 0.057** (0.026)

0.077 (0.047) 0.114*** (0.040) 0.002 (0.026) 0.042 (0.068) 0.040 (0.029)

Semi skilled, 1860 Farmer, 1860 Unskilled blue collar, 1860 No occupation, 1860 Age & Birthplace FE's 1860 county FE's Observations R-squared

(5)

-0.007 (0.019) -0.012 (0.009) 0.023 (0.028) 0.070 (0.043) 0.033 (0.036) -0.002 (0.023) 0.061 (0.061) 0.055** (0.026)

Y N

Y Y

Y Y

Y Y

Y N

Y Y

Y Y

Y Y

3,693 0.043

3,693 0.100

2,415 0.158

3,513 0.137

3,693 0.102

3,693 0.152

2,415 0.205

3,513 0.161

Table 7: Impact of County Characteristics on Migration Propensity (1) Dependent variable: Social indicator Union Union X Social indicator, 1860

(3) (4) (5) Moved counties, 1860-80 1860 Douglas vote share 1860 % enslaved

-0.089* (0.046) 0.247** (0.098)

White collar, 1860 Semi skilled, 1860 Farmer, 1860 Unskilled blue collar, 1860 No occupation, 1860 White collar X Social indicator, 1860 Semi skilled X Social indicator, 1860 Farmer X Social indicator, 1860 Unskilled blue collar X Social indicator, 1860 No occupation X Social indicator, 1860

(2)

-0.079* (0.047) 0.226** (0.098) 0.039 (0.094) 0.146* (0.082) -0.022 (0.053) -0.090 (0.131) 0.048 (0.050) 0.114 (0.242) -0.057 (0.231) 0.108 (0.114) 0.450 (0.357) 0.048 (0.108)

Log family wealth, 1860

-0.031 (0.036) 0.249 (0.160)

-0.028 (0.036) 0.242 (0.160) 0.153* (0.079) 0.170** (0.067) 0.049 (0.039) 0.080 (0.116) 0.079* (0.041) -0.405 (0.354) -0.181 (0.300) -0.166 (0.198) -0.195 (0.560) -0.048 (0.192)

-0.039* (0.022) -0.031 (0.052)

Log family wealth X Social indicator, 1860 Observations R-squared

-0.091* (0.047) 0.186* (0.100)

3,693 0.102

3,693 0.108

27

3,514 0.134

(6)

-0.071** (0.036) 0.347** (0.166)

-0.079*** (0.018) 0.142** (0.070) 3,693 0.101

3,693 0.107

3,514 0.135

Table 8: Locational Choices of Interstate Migrants (1)

(2)

(3)

(4) Migrated to:

-0.098*** (0.034) 0.005 (0.065) -0.001 (0.057) -0.031 (0.041) 0.093 (0.085) -0.058 (0.045) -0.001 (0.016) 0.017 (0.054)

-0.027 (0.029) 0.027 (0.054) 0.006 (0.048) -0.025 (0.034) 0.089 (0.071) -0.024 (0.038) 0.004 (0.013) -0.008 (0.045) -0.255*** (0.014) -1.408*** (0.099)

-0.004 (0.018) -0.052 (0.034) -0.030 (0.030) 0.002 (0.021) 0.087* (0.045) -0.011 (0.023) 0.001 (0.008) -0.064** (0.028)

-0.024 (0.017) -0.072** (0.033) -0.033 (0.029) 0.003 (0.020) 0.090** (0.042) -0.017 (0.023) -0.002 (0.008) -0.057** (0.027) 0.081*** (0.008) 0.187*** (0.059)

1,053 0.287

1,050 0.508

1,053 0.425

1,050 0.481

Dependent variable: South Union Family wealth, 1860 Literate, 1860 White collar, 1860 Semi-skilled, 1860 Farmer, 1860 Unskilled blue collar, 1860 No occupation, 1860 County log farm value per acre, 1880 County % farms owned, 1880 Observations R-squared

(5)

(6)

(7)

0.148*** (0.039) -0.044 (0.073) 0.045 (0.065) 0.035 (0.046) -0.176* (0.097) 0.060 (0.051) -0.009 (0.018) 0.077 (0.061)

0.103*** (0.037) -0.065 (0.071) 0.038 (0.062) 0.029 (0.044) -0.174* (0.092) 0.037 (0.049) -0.012 (0.017) 0.092 (0.058) 0.163*** (0.018) 0.833*** (0.128)

-0.048*** (0.018) 0.089*** (0.033) -0.012 (0.030) -0.010 (0.021) -0.003 (0.044) 0.006 (0.023) 0.007 (0.008) -0.031 (0.028)

-0.054*** (0.017) 0.110*** (0.033) -0.010 (0.029) -0.011 (0.021) -0.006 (0.043) 0.002 (0.023) 0.008 (0.008) -0.029 (0.027) 0.005 (0.009) 0.381*** (0.060)

1,053 0.276

1,050 0.344

1,053 0.147

1,050 0.188

Northeast

(8)

Midwest

West

Note: All regressions contain age, birthplace and 1860 county fixed effects. Sample consists of interstate migrants.

Table 9: Locational Choices of Intrastate Migrants (1) Dependent variable:

Union

(3)

(4)

1860 % enslaved in 1880 county of residence

-0.044** (0.018)

-0.036* (0.019) 0.055** (0.025) 0.009 (0.010) -0.043 (0.055) -0.073** (0.032) -0.006 (0.021) 0.021 (0.053) 0.007 (0.023)

-0.023** (0.011)

-0.020* (0.011) -0.001 (0.016) 0.016*** (0.006) 0.010 (0.035) -0.014 (0.022) 0.003 (0.014) -0.007 (0.040) 0.014 (0.015)

787 0.432

727 0.453

787 0.418

727 0.446

Family wealth, 1860 Literate, 1860 White collar, 1860 Semi-skilled, 1860 Farmer, 1860 Unskilled blue collar, 1860 No occupation, 1860 Observations R-squared

(2)

1860 vote share to Douglas in 1880 county of residence

Note: All regressions contain age, birthplace, and 1860 county fixed effects. Sample consists of internal migrants in Kentucky

28

Table 10: Selection of Union and Confederate Migrants (1) Dependent variable: Sample: Family wealth 1860 Literate 1860 Log occupational score 1860

Union -0.039** (0.018) 0.010 (0.036) 0.093*** (0.034)

(2)

(3) (4) Moved counties, 1860-1880 Confederate p (u=c) Union Confederate -0.060*** (0.023) -0.000 (0.091) 0.055 (0.064)

0.472 0.920

Semi skilled, 1860 Farmer, 1860 Unskilled blue collar, 1860 No occupation, 1860 1,862 0.172

-0.047*** (0.013) -0.010 (0.034)

-0.041** (0.018) -0.031 (0.081)

0.772

0.124** (0.058) 0.121*** (0.045) 0.014 (0.030) 0.019 (0.076) 0.027 (0.033)

-0.102 (0.097) 0.023 (0.101) -0.047 (0.056) 0.174 (0.168) 0.035 (0.064)

0.048

2,716 0.147

797 0.302

0.817

0.611

White collar, 1860

Observations R-squared

p (u=c)

553 0.370

Note: All regressions contain birthplace and 1860 county fixed effects.

29

0.379 0.338 0.407 0.917

Table 11: Evidence of Di↵erential Migrant Selection by County

Dependent variable Sample Log occupation score, 1860 X Douglas vote share, 1860 Log family wealth, 1860 X Douglas vote share, 1860

(1) (2) Migrated counties Union Confederate 0.178 (0.175) 0.144 (0.094)

-0.609** (0.309) -0.046 (0.115)

p(u=c)

0.206

Moved counties

Log family wealth, 1860

30 Observations R-squared Log occupation score, 1860 X % Slave, 1860 Log family wealth, 1860 X % Slave, 1860

0.095*** (0.035) -0.037** (0.018)

0.087 (0.066) -0.055** (0.023)

1,862 0.173

553 0.376

-0.339 (0.275) 0.292** (0.127)

0.191 (0.465) 0.022 (0.164)

0.913 0.532

Log family wealth, 1860 Observations R-squared

0.045 (0.069) -0.062** (0.028)

1,862 0.175

553 0.370

-0.133 (0.183) -0.005 (0.042) 0.389*** (0.050) 0.018 (0.018)

1,816 0.262

539 0.454

-0.310** (0.154) 0.028 (0.018) 0.288*** (0.025) 0.040*** (0.013)

0.230 (0.275) -0.024 (0.040) 0.391*** (0.050) 0.018 (0.018)

1,816 0.260

539 0.454

0.055 0.347 0.06 0.352

0.196

Moved counties 0.098*** (0.034) -0.049*** (0.018)

0.245*** (0.089) 0.037** (0.018) 0.287*** (0.025) 0.038*** (0.013)

0.686

Moved counties X % Slave, 1860

Log occupational score, 1860

p(u=c)

0.029

Moved counties X Douglas vote share, 1860

Log occupational score, 1860

(3) (4) Log occupational score, 1880 Union Confederate

0.498 0.711

0.077 0.216 0.057 0.317

Table 12: Migration Behavior of Brothers of Civil War Recruits (1) Dependent variable Union brother Union brother X Douglas vote share, 1860

0.025 (0.028)

2,195 0.140

Dependent variable Sample

Union 0.032 (0.021) 0.210** (0.101)

2,195 0.141

0.014 (0.014)

524 0.366

518 0.415

Log occupational score, 1880 Confederate p(u=c) Union 0.035 (0.052) 0.064 (0.207)

Moved counties X % Slave, 1860 Observations R-squared

-0.044* (0.025)

0.345 (0.222)

Observations R-squared

Moved counties X Douglas vote share, 1860

(3) (4) Destination county: Douglas vote % Slave share

Moved counties 0.034 (0.028) -0.044 (0.124)

Union brother X % Slave, 1860

Moved counties

(2)

1,613 0.245

416 0.371

Note: Sample includes younger brothers (ages 0-9 in 1860) of Union and Veteran recruits

31

0.948

Confederate

p(u=c)

0.025 (0.021)

0.039 (0.049)

0.791

-0.194 (0.192)

0.138 (0.398)

0.438

1,613 0.244

416 0.371

0.146

Figure 1: Changes in Distribution of Southern Born, 1850-1890

−2

Percentage points −1 0 1

2

Change in % of Southern Born Men From Last Census Year

1860

1870

1880

1890

Year Living in Northeast Living in South

Living in Midwest Living in West ®

Source: Haines & ICPSR (2010).

32

Figure 2: Geographic Distribution of Union and Confederate Soldiers, 1860

33

Figure 3: Age Distribution of Union and Confederate Soldiers, 1860

0

.02

Density .04

.06

.08

Age Distribution of Union and Confederate Recruits, 1860

10

20

30 Age Union

40

50

Confederate ®

34

Figure 4: Distribution of Migrant Union and Confederate Veterans, 1880

35